City-wide Traffic Volume Inference with Loop Detector Data and Taxi Trajectories Chuishi Meng 1 , Xiuwen Yi 2 , Lu Su 1 , Jing Gao 1 , Yu Zheng 2 1. University at Buffalo, State University of New York 2. Urban Computing Group, Microsoft Research, Beijing China ACM SIGSPATIAL 2017
28
Embed
City-wide Traffic Volume Inference with Loop Detector Data ... · City-wide Traffic Volume Inference with Loop Detector Data and Taxi Trajectories Chuishi Meng 1, Xiuwen Yi2, Lu Su
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
City-wide Traffic Volume Inference with
Loop Detector Data and Taxi Trajectories
Chuishi Meng1, Xiuwen Yi2, Lu Su1, Jing Gao1, Yu Zheng2
1. University at Buffalo, State University of New York
2. Urban Computing Group, Microsoft Research, Beijing China
ACM SIGSPATIAL 2017
Urban Transportation Challenges
Traffic
Congestion
Parking
Difficulty
Longer
Commuting
Environment
& Energy
Traffic Volume
• Definition– Total number of vehicles traversing by a road segment
during a time window
• A unique traffic condition metric– Most common measurement is Travel Speed
– The volume reveals detailed condition information
besides average speed
• Applications
Speed-based Traffic Conditions
Traffic Control Pollution Emission
Loop Detectors
• Loop Detectors
– Sensors buried under the pavements, can detect
vehicles passing by
• Pro
– Accurate
• Con– Expensive & not scalable
– Extreme sparsity (155/19165 in Guiyang)
Taxi Trajectories• GPS Trajectory
– A sequence of time-ordered spatial points
• Pro– High coverage
• Con– Only a biased sample of all vehicles
– no direct information about total volume
Loop Detectors & Taxi Trajectories
• High Accuracy
• Low Coverage
• High Coverage
• Low Accuracy
Loop Detectors & Taxi Trajectories
• High Accuracy
• Low Coverage
• High Coverage
• Low Accuracy
Goal
• Infer city-wide traffic volume on each road segment based on
– Loop detector data
– Taxi trajectories
– Urban context
Taxi Trajectories Citywide Traffic Volume
Road Network
Meteorology
Loop Detectors
POI
Main Idea
• Graph-based Semi-supervised learning– Take advantage of the benefits of both data sources
– Construct traffic affinity graph with taxi trajectories
– Estimate city-wide traffic volume with loop detector data
– High coverage & High Accuracy
• Incorporate spatio-temporal properties of traffic volume– Constructing the road affinity graph
Road segment with loop detector Road segment w/t loop detector
Spatial Edge Temporal EdgeTemporal Layers
…
Framework
Loop Detector
Data
Road
Network
Taxi
Trajectories
Meteorology
Data
City-wide Travel
Speed Estimation
ST-SSL Model
LearningPoint of
Interests
Inference Confidence
Estimation
Road Daily
Speed Pattern
Speed Pattern
Extraction
Instant Traffic
Volume
Traffic Volume
Aggregator
Context Extractor
Road Features
POI FeaturesAffinity Graph
Construction
Affinity Graph Edge
Weight Learning
Meteorology
Features Instant Travel
Speed
Main Procedures
• Affinity Graph Construction– Determine the graph structure
• Graph Edge Weight Learning– Learn the correlations with urban context
• Spatio-temporal Semi-supervised Learning– Estimate traffic volume of every road segment
𝒕𝒊𝒕𝒊−𝟏𝒕𝒊−𝟐𝒕𝒊−𝟑𝒕𝒊−𝒄
Current LayerRecent LayersPeriodical Layers
……
with loop detector
w/t loop detector
Temporal Layers
Framework
Loop Detector
Data
Road
Network
Taxi
Trajectories
Meteorology
Data
City-wide Travel
Speed Estimation
ST-SSL Model
LearningPoint of
Interests
Inference Confidence
Estimation
Road Daily
Speed Pattern
Speed Pattern
Extraction
Instant Traffic
Volume
Traffic Volume
Aggregator
Context Extractor
Road Features
POI FeaturesAffinity Graph
Construction
Affinity Graph Edge
Weight Learning
Meteorology
Features Instant Travel
Speed
Preliminary: City-wide Travel Speed Estimation
• Context-Aware Tensor Decomposition* (CATD)
• Output: travel speed on every road segment
* Wang, Yilun, Yu Zheng, and Yexiang Xue. "Travel time estimation of a path using sparse trajectories." Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2014.
Framework
Loop Detector
Data
Road
Network
Taxi
Trajectories
Meteorology
Data
City-wide Travel
Speed Estimation
ST-SSL Model
LearningPoint of
Interests
Inference Confidence
Estimation
Road Daily
Speed Pattern
Speed Pattern
Extraction
Instant Traffic
Volume
Traffic Volume
Aggregator
Context Extractor
Road Features
POI FeaturesAffinity Graph
Construction
Affinity Graph Edge
Weight Learning
Meteorology
Features Instant Travel
Speed
Traffic Affinity Graph Construction
• Affinity Graph– Node: per-lane volume on a road at one time slot
– Edge: traffic condition similarity between two nodes
• Edge Construction with spatio-temporal knowledge– Spatial: connect to segments with similar traffic patterns
• within a radius (e.g., 200m)
• within several hops in the road network
• with several similar roads equipped loop detectors
• satisfies: 1) have similar average speed, 2) have similar daily speed patterns (characterized by Pearson correlations)
– Temporal: connect to correlated temporal layers
• connect to recent temporal layers
• connect to periodical temporal layers (day/week)